Systematic lowering of the scaling of Monte Carlo calculations by partitioning andsubsampling
Authors:
Antoine Bienvenu,
Jonas Feldt,
Julien Toulouse,
Roland Assaraf
Abstract:
We propose to compute physical properties by Monte Carlo calculations using conditional expectation values. The latter are obtained on top of the usual Monte Carlo sampling by partitioning the physical space in several subspaces or fragments, and subsampling each fragment (i.e., performing side walks) while freezing the environment. No bias is introduced and a zero-variance principle holds in the…
▽ More
We propose to compute physical properties by Monte Carlo calculations using conditional expectation values. The latter are obtained on top of the usual Monte Carlo sampling by partitioning the physical space in several subspaces or fragments, and subsampling each fragment (i.e., performing side walks) while freezing the environment. No bias is introduced and a zero-variance principle holds in the limit of separability, i.e. when the fragments are independent. In practice, the usual bottleneck of Monte Carlo calculations -- the scaling of the statistical fluctuations as a function of the number of particles N -- is relieved for extensive observables. We illustrate the method in variational Monte Carlo on the 2D Hubbard model and on metallic hydrogen chains using Jastrow-Slater wave functions. A factor O(N) is gained in numerical efficiency.
△ Less
Submitted 3 August, 2022; v1 submitted 2 May, 2022;
originally announced May 2022.
Efficient treatment of local meta-generalized gradient density functionals via auxiliary density expansion: the density fitting (DF) J+X approximation
Authors:
Alyssa V. Bienvenu,
Gerald Knizia
Abstract:
We report an efficient technique to treat density functionals of the meta-generalized gradient approximation (mGGA) class in conjunction with density fitting of Coulomb terms (DF-J) and exchange-correlation terms (DF-X). While the kinetic energy density $τ$ cannot be computed in the context of a DF-JX calculation, we show that the Laplacian of the density $\upsilon$ can be computed with almost no…
▽ More
We report an efficient technique to treat density functionals of the meta-generalized gradient approximation (mGGA) class in conjunction with density fitting of Coulomb terms (DF-J) and exchange-correlation terms (DF-X). While the kinetic energy density $τ$ cannot be computed in the context of a DF-JX calculation, we show that the Laplacian of the density $\upsilon$ can be computed with almost no extra cost. With this technique, $\upsilon$-form mGGAs become only slightly more expensive (10%--20%) than GGAs in DF-JX treatment---and several times faster than regular $τ$-based mGGA calculations with DF-J and regular treatment of the density functional. We investigate the translation of $\upsilon$-form mGGAs into $τ$-form mGGAs by employing a kinetic energy functional, but find this insufficiently reliable at this moment. However, $\upsilon$ and $τ$ are believed to carry essentially equivalent information beyond $ρ$ and $\Vert\vec\nablaρ\Vert$ [Phys. Rev. B 2007, 75, 155109], so a reparametrization of accurate mGGAs from the $τ$-form into the $\upsilon$-form should be possible. Once such functionals become available, we expect the presented technique to become a powerful tool in the computation of reaction paths, intermediates, and transition states of medium sized molecules.
△ Less
Submitted 5 February, 2018; v1 submitted 27 October, 2017;
originally announced October 2017.
